| Re: [aspectj-users] Interesting Problem & Reweavable |
|
I second Ron and Nick. Seems like there is already a bug to fix this issue: https://bugs.eclipse.org/bugs/show_bug.cgi?id=91417 For an additional use case, see: https://bugs.eclipse.org/bugs/show_bug.cgi?id=36430 -Ramnivas Lesiecki Nicholas wrote: I agree that reweavable should be the default to enable precisely the situation Ron describes. Cheers, Nick --- Ron Bodkin <rbodkin@xxxxxxxxxxxxxx> wrote:Hi All, I've been working on applying the AspectJ 5 load-time weaving support from the branch that Alex and Jonas have been working on, to let me weave some monitoring code into an app server at load time. The code is still fairly new, but I was able to get it working for my environment, which was great (already much better than trying to weave into a wad of 3rd party jars at build time!). Great work guys! I plan to blog about this and to submit a few small suggestions to bugzilla and aspectj-dev. However, I ran into the following output when I was weaving into the MySQL JDBC driver's debug version: error at com\mysql\jdbc\NonRegisteringDriver.java::0 class 'com.mysql.jdbc.NonRegisteringDriver' is already woven and has not been built with -Xreweavable That's right. This error means that the code was already woven with AspectJ! The MYSQL debug JDBC driver includes a tracing aspect: it's great to see this kind of increased use & popularity of AspectJ. For the curious, you can see the source code for this aspect in the source distribution of the Mysql Connector/J. In this case, the normal, non-debug, library doesn't use AspectJ so it's not a major problem.(*) Unfortunately, because the default today is to not emit reweavable bytecode, it makes it challenging to weave into libraries like this. I think the increasing amount of AspectJ use in libraries makes it's more important than ever to make -Xreweavable the default option (or the compressed version). Indeed, under what circumstances would we not want code to be reweavable? Is the performance that much better? It seems to me that maybe reweavable should just always be the case. Failing that, I would like to see an option to weave into previously non-reweavable woven code at a bytecode-level, which would just treat the woven result like one would treat any Java bytecode (i.e., it wouldn't recognize adviceexecution, nor understand ITD's, etc., it would operate on the "machine-level" translation into Java bytecode). Ron Ron Bodkin Chief Technology Officer New Aspects of Software w: (415) 824-4690 (*) Amusingly, I only discovered this because I had an error in the first zip distribution I downloaded so I could only extract the debug jar file. _______________________________________________ aspectj-users mailing list aspectj-users@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/aspectj-users_______________________________________________ aspectj-users mailing list aspectj-users@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/aspectj-users |